Spectral unmixing techniques strive to find proportions of end-members within a pixel from the observed mixed pixel spectrum and a number of pure end-member spectra of known composition. The outcomes of such analysis are fraction (abundance) images for the selected (pure) end-members and a root mean square (RMS) error estimate representing the difference between the observed mixed spectrum and the calculated mixed spectrum. The RMS image can be used to select additional end-members and re-position existing ones. This is now done manually. In this Letter, an automatediterative approach is proposed using the RMS error image to select additional end-members and re-distribute older ones in order toincrease the accuracy of the spectral unmixing. Optimization criteria are proposed to drive the iterative process including minimization of the average RMS, minimizing the spread of the RMS values, minimizing the spatial structure of the RMS image, minimizing the spatial anisotropy of the RMS image and minimizing the local variance. The preliminary results of the analysis indicate that considerable improvement tothe spectral unmixing results are achieved using the iterative spectral unmixing (ISU) approach.